Vectorize MCP Server

npx -y @vectorize-io/vectorize-mcp-server@latest

{
  "mcpServers": {
    "vectorize-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@vectorize-io/vectorize-mcp-server@latest"
      ]
    }
  }
}

Three tools, and the second one is the reason to keep this server around even outside retrieval work. retrieve does vector search against a pipeline with a configurable k. extract takes a base64 document plus its content type and returns text already chunked into Markdown, which means an agent can turn a PDF into usable context without that file ever having been ingested into the pipeline. deep-research generates a private research report from the pipeline's own corpus, optionally widened with web search. Configuration is three environment variables — organisation id, token and pipeline id — so the pipeline is fixed per server instance rather than chosen per call, and pointing at a second corpus means running a second instance.